Neural networks ought to be very appealing to hackers. You can easily implement them in hardware or software and relatively simple networks can perform powerful functions. As the jobs we ask of neural networks get more complex, the networks require more artificial neurons. That’s why researchers are pursuing dense integrated neuron chips that could do for neural networks what integrated circuits did for conventional computers.
Researchers at Princeton have announced the first photonic neural network. We recently talked about how artificial neurons work in conventional hardware and software. The artificial neurons look for inputs to reach a threshold which causes them to “fire” and trigger inputs to other neurons.
To map this function to an optical device, the researchers created tiny circular waveguides in a silicon substrate. Light circulates in the waveguide and, when released, modulates the output of a laser. Each waveguide works with a specific wavelength of light. This allows multiple “inputs” (in the form of different wavelengths) to sum together to modulate the laser.
The team used a 49-node network to model a differential equation. The photonic system was nearly 2,000 times faster than other techniques. You can read the actual paper online if you are interested in more details.
There’s been a lot of work done lately on both neural networks and optical computing. Perhaps this fusion will advance both arts.
I imagine somewhere in the paper just how flexible this design is to being modified.
This story gets an instant dose of skepticism from me for just mentioning photonics. We’ve had a hard enough time building transistors, and you expect me to believe you’ve just figured out how to build sprawling neural nets capable of modeling differential equations? (are you sure the differential equation wasn’t written to proximate the neural network instead?)
Either this is absolute baloney, baloney enough insofar as suffering from horrible gate fan-out and such, or we might be about to finally make some progress in regular old photonic computing.
photonic transistors*
and photonic neural nets*
approximate**
FEATURE REQUEST: EDIT BUTTON
The problem with an edit button would be people changing things to make others look stupid or themselves look less so. Unless the changes had to be approved by staff to ensure it’s only minor corrections…
You’re right. People would do that. There would need to be an edit history available.
Force the edits to appear inline, maybe?
The problem with an edit button in the Hackaday comments is that WordPress is a complete fucking mess, we’ve looked into it, and giving commenters an ‘edit’ ability is number eight thousand on the list of features we want.
What do we *actually* need? Integration with hackaday.io users. Voting, so the idiotic, toxic, nincompoopery will fall to the bottom of the comments thread. Better CSS, because you can clearly see this comment chain is fucked after four replies. We need WordPress VIP to stop sucking, and we need a reason to actually *do* this.
Making Hackaday Comments Great Again™ is going to cost time and money, and it’s very doubtful anything we do would be worth what we would get in return. This is, after all, one of the most toxic places on the Internet. I believe it is beyond help, and any effort to improve the commenting system would only result in more trolling. That would be *after* the complaints we would get that *horrors, the comment system is changing*.
EDIT, because I can edit and you can’t. Take a look at this. It’s a thread about Queercon, last DEF CON. That’s also a comment thread questioning why people should have the right to assemble. I recognize the word ‘fascism’ has been thrown around a lot lately, but some Hackaday commenters are actual fascists. These commenters should die – slowly – in a fire. I would be more than happy to shut the whole thing down before giving these jackasses a better platform to spew their bile.
http://www.wpbeginner.com/plugins/allow-users-edit-comments-wordpress/
havent there been several functional prototypes?
as far as i know it has always been scaling that has been the issue both in an economical and size sense.
that said i read the synopsis and it does say that it emulates the differential equation, what that actually means in practice i dont know.
The device is an emulator for neural networks, and the network is defined by the data within it therefore it is a general purpose device?